Sentinel-2 eo:cloud_cover versus s2:high_proba_clouds_percentage

[Scott Henderson (scottyhq)]

Just documenting something that may already be well-known by regular sentinel-2 users, but I found surprising, which is that the eo:cloud_cover estimate can be very poor in some regions. eo:cloud_cover is standard in STAC so it's a reasonable starting place for workflows, including the PC S2 L2a Example Notebook

It looks like eo:cloud_cover is copied from CLOUDY_PIXEL_PERCENTAGE in the original metadata, but there are other cloud probability classes like s2:high_proba_clouds_percentage in STAC metadata which may be better suited for filtering searches in certain areas (MGRS 45RVM shown below) .

You might accidentally omit a lot of good scenes depending on which of these fields you use!

total scenes: 408
# scenes 'eo:cloud_cover'<10: 35
# scenes 's2:medium_proba_clouds_percentage: 120
# scenes 's2:high_proba_clouds_percentage'<10: 191

import pystac_client
import planetary_computer
import geopandas as gpd
from IPython.display import Image

catalog = pystac_client.Client.open(
    "https://planetarycomputer.microsoft.com/api/stac/v1",
)

search = catalog.search(
    collections=["sentinel-2-l2a"],
    bbox=[86.5, 28.0, 87.0, 28.5],
    datetime=['2015-01-01','2023-01-01'],
    query={"s2:mgrs_tile": {"eq":"45RVM"},
          },
    sortby = 'datetime',
)

items = search.get_all_items()
gf = gpd.GeoDataFrame.from_features(items.to_dict(), crs='EPSG:4326')
gf['stac_id'] = [item.id for item in items]
print(f"total scenes: {len(gf)}")
print(f"# scenes 'eo:cloud_cover'<10: {(gf['eo:cloud_cover']<10).sum()}")
print(f"# scenes 's2:medium_proba_clouds_percentage: {(gf['s2:medium_proba_clouds_percentage']<10).sum()}")
print(f"# scenes 's2:high_proba_clouds_percentage'<10: {(gf['s2:high_proba_clouds_percentage']<10).sum()}")

i = 1
s = gf.iloc[i]
print(s['stac_id'])
print('eo:cloud_cover=',s['eo:cloud_cover'])
print('s2:medium_proba_clouds_percentage=',s['s2:medium_proba_clouds_percentage'])
print('s2:high_proba_clouds_percentage=',s['s2:high_proba_clouds_percentage'])

Image(url=items[i].assets['rendered_preview'].href, width=400)